Moonshot AI released Kimi K3.1 Pro on October 23, 2026, and the numbers say it is now the strongest open-weight AI model money can buy. On Terminal-Bench 4.0 — one of the toughest tests for AI coding agents — the Chinese lab’s new model scored 65.7%, landing within a single point of Anthropic’s Claude Opus 5.5 (66.4%) and well ahead of OpenAI’s GPT-6 Astra (58.2%). The full model weights are promised for release by November 4, and the API costs a third less than its predecessor.
What happened
Kimi K3.1 Pro is the first model in Moonshot’s new K3.1 series. It is not a brand-new base model: Moonshot describes it as a heavily post-trained version of Kimi K3, keeping the same architecture — roughly 2.8 trillion total parameters with 104 billion active per token via a mixture-of-experts design, handling text and image input — but run through far more reinforcement learning across what the company calls “hundreds of downstream domains.”
The headline specs:
- 1-million-token context window — enough to load an entire codebase or a stack of long documents in a single pass.
- Two flavors at launch: Kimi K3.1 Pro (standard) and Kimi K3.1 Pro Highspeed, which Moonshot says runs two to three times faster for roughly double the price. Both are live now in the Kimi app, Kimi Work, Kimi Code, and the Kimi API under the identifiers
kimi-k3.1-proandkimi-k3.1-pro-highspeed. - API pricing cut by a third: $2 per million input tokens and $10 per million output tokens (down from K3’s $3/$15), with cached input at $0.20. Highspeed costs $4/$20.
- Open weights due by November 4 — no license has been specified yet.
Moonshot’s own benchmark table is aggressive: top of Terminal-Bench 4.0’s open tier, best score on the company’s OfficeQA Pro v2 office-work test, and a claim-level hallucination rate that fell from 23.4% on K3 to 8.3%. Independent benchmarker Artificial Analysis ranks it fourth on its Intelligence Index with 55 points — behind only Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. An important caveat applies to all of this: Moonshot’s numbers are unaudited, and much of its coding evaluation ran inside Anthropic’s Claude Code harness. Artificial Analysis also notes the model is “very wordy.”
One detail stands out: Moonshot says it deliberately stripped cyberattack-related training data out of the model so the open release won’t push the frontier of offensive hacking.
Why it matters for SMBs
For a small business, the frontier model race is mostly a pricing story. A coding agent near the level of Claude Opus 5.5 at $2/$10 per million tokens — versus flagship-model prices that can run many times higher — changes the math on everyday work: reviewing a vendor contract, refactoring an internal tool, building a reporting dashboard, or drafting a long technical document.
The 1-million-token context window is the real unlock. Feed an agent your entire product catalog, last quarter’s support tickets, and your SOP library in one session, and it reasons across all of it instead of forgetting page three. That is exactly the use case AI coding assistants keep improving at — and Kimi now does it at a fraction of the cost.
The open-weights promise matters too. Once weights ship on November 4, teams with capable hardware — or a friendly IT provider — can run the model privately, with no per-token billing and no data leaving their machines. For businesses handling client data, that is a meaningful option to have.
Pricing, availability, and what to do next
- Price: Kimi API — $2 input / $10 output per million tokens ($0.20 cached input); Highspeed $4/$20. That is a 33% cut from Kimi K3.
- Availability: live now in the Kimi app, Kimi Work desktop app, Kimi Code, and the Kimi API (also inside Codex). Full open weights promised by November 4, 2026 — license terms still unannounced.
- What to do: if your team already pays for a frontier coding assistant, run a side-by-side trial this week — the cost delta is large enough to be worth an afternoon of testing. If you are on free AI tools, the Kimi app is the zero-risk way to try a near-frontier model before touching the paid API.
Sources
- Skimwire — Moonshot’s Kimi K3.1 Pro ranks fourth on a major AI index, with open weights due Nov. 4
- HandyAI “Model Drop” — Kimi K3.1 Pro (ship date October 23, 2026; pricing $2/$10 per M tokens)
- MadRobot — Kimi K3.1 Pro nearly matches Claude Opus 5.5, weights by Nov 4
- Startup Fortune — Moonshot AI releases Kimi K3.1 Pro open weight model to rival Claude
- Toto’s AI Diary — AI news on October 23, 2026: Moonshot releases Kimi K3.1 Pro
